🎉 "Harbor Adapters and Harbor-Index" has been accepted to NeurIPS 2026 (E&D)
Harbor Adapters and Harbor-Index:Infrastructure and Curated Meta-Dataset for Large-Scale Agentic Evaluation. Work with Harbor Adapter team
Professional Summary
I am a postdoctoral associate at Boston University, working with Prof. Najoung Kim and Prof. Sebastian Schuster. I received my Ph.D. from Korea University, advised by Prof. Pilsung Kang. During my Ph.D, I was research intern at NAVER and contributed to CLOVA for Writing. I completed my B.S. at HUFS, advised by Prof. Chungmok Lee
My research focuses on LLM evaluation, aiming to discover, measure, understand, and improve the capabilities of language models. My long-term research vision is to establish a science of evaluation for language models. I like to think about what makes evaluation reliable and what it truly tells us about these models. I am also interested in LLM agents that autonomously solve complex problems in research and engineering, and how to reliably evaluate them.
Education
PhD Industrial Management & Engineering
Korea University
BSc Industrial Management & Engineering, BA International Finance (Double Major)
HUFS
Interests
Language Model Evaluation
LLM Agent
Math/Code Reasoning
Writing with AI
Anomaly Detection
Recent News See all
🎉 Three papers accepted at EMNLP 2026 Main conference!
➊ AI Writers Have a Consistent Stylometric Footprint, but AI Editors Do Not ➋ PGMem: Tightly Coupled Persona–Memory Graph for Lifelong Personalized Agents and ➌ LLMs Learn Better In-Context from Rules than from Examples
💡 Check out our math reasoning paper "Cliff Token"
‘Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning’ now available on arXiv! Work with Jaeyong Ko and Pilsung Kang. Congrats to Jaeyong on his very first paper 🎉
💡 Check out our implicit CoT paper "CIRF"
‘CIRF: Tokenizing Chain-of-Thoughts into Reusable Functional Units for Efficient Latent Reasoning in Large Language Models’ now available on arXiv! Collaborated with HUFS team 🦉
🎉 "Can Structural Cues Save LLMs?" has been accepted to KDD 2026!
Can Structural Cues Save LLMs? Evaluating Language Models in Massive Document Streams. Work with Yebin Lim, Woojun Jung, Wonjun Choi and Prof. Susik Yoon 🐯
Selected Publications See all
For an up-to-date list of publications, check out my Google Scholar.
(* denotes equal contribution, † denotes equal contribution as senior role.)
CheckEval: A reliable LLM-as-a-Judge framework for evaluating text generation using checklists
EMNLP 2025·Nominated for Outstanding Paper Award